Use of G-methods for handling time-varying confounding in observational research.